https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96338
Bug ID: 96338
Summary: [SVE] Unnecessary register saves in exception handler
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: rtl-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: rsandifo at gcc dot gnu.org
Target Milestone: ---
Compiling the following testcase with -march=armv8.2-a+sve -O2:
---------------------------------------------------------------------
void bar (__SVFloat32_t);
void
foo (__SVFloat32_t x)
{
try { bar (x); } catch (...) { bar (x); throw; }
}
---------------------------------------------------------------------
gives an EH handler like the following;
---------------------------------------------------------------------
.cfi_restore_state
str z8, [sp, #2, mul vl]
.cfi_escape 0x10,0x48,0x8,0x8f,0,0x92,0x2e,0,0x40,0x1e,0x22
str z9, [sp, #3, mul vl]
.cfi_escape 0x10,0x49,0x8,0x8f,0,0x92,0x2e,0,0x48,0x1e,0x22
str z10, [sp, #4, mul vl]
.cfi_escape 0x10,0x4a,0x9,0x8f,0,0x92,0x2e,0,0x8,0x20,0x1e,0x22
str z11, [sp, #5, mul vl]
.cfi_escape 0x10,0x4b,0x9,0x8f,0,0x92,0x2e,0,0x8,0x28,0x1e,0x22
str z12, [sp, #6, mul vl]
.cfi_escape 0x10,0x4c,0x9,0x8f,0,0x92,0x2e,0,0x8,0x30,0x1e,0x22
str z13, [sp, #7, mul vl]
.cfi_escape 0x10,0x4d,0x9,0x8f,0,0x92,0x2e,0,0x8,0x38,0x1e,0x22
str z14, [sp, #8, mul vl]
.cfi_escape 0x10,0x4e,0x9,0x8f,0,0x92,0x2e,0,0x8,0x40,0x1e,0x22
str z15, [sp, #9, mul vl]
.cfi_escape 0x10,0x4f,0x9,0x8f,0,0x92,0x2e,0,0x8,0x48,0x1e,0x22
str z16, [sp, #10, mul vl]
str z17, [sp, #11, mul vl]
str z18, [sp, #12, mul vl]
str z19, [sp, #13, mul vl]
str z20, [sp, #14, mul vl]
str z21, [sp, #15, mul vl]
str z22, [sp, #16, mul vl]
str z23, [sp, #17, mul vl]
str p5, [sp, #1, mul vl]
str p6, [sp, #2, mul vl]
str p7, [sp, #3, mul vl]
str p8, [sp, #4, mul vl]
str p9, [sp, #5, mul vl]
str p10, [sp, #6, mul vl]
str p11, [sp, #7, mul vl]
str p12, [sp, #8, mul vl]
str p13, [sp, #9, mul vl]
str p14, [sp, #10, mul vl]
str p15, [sp, #11, mul vl]
str p4, [sp]
bl __cxa_begin_catch
add x0, sp, 32
ldr z0, [x0, #18, mul vl]
.LEHB1:
bl _Z3bar13__SVFloat32_t
bl __cxa_rethrow
.LEHE1:
.L5:
mov x19, x0
bl __cxa_end_catch
mov x0, x19
.LEHB2:
bl _Unwind_Resume
---------------------------------------------------------------------
The spills of z8-z23 and p4-p15 are completely unnecessary,
since the noreturn _Unwind_Resume call ensures that this code
never returns to foo's caller.
I think the fix is to make ira.c:update_equiv_regs_prescan
a bit smarter: if a call occurs in a block that has no
path to the exit block, then we only need to consider
(parts of) registers that are clobbered by the callee
but preserved by both eh_edge_abi and the current
function's abi.
If, as is usual, eh_edge_abi is the lowest common denominator
among the available ABIs, no callees will clobber something
that is preserved by eh_edge_abi.
